Latest / Wired to Build / AI Doesn't Know What Good Looks Like. You Do. | Marcus Turner @ Constructrr

Transcript

AI Doesn't Know What Good Looks Like. You Do. | Marcus Turner @ Constructrr · Wired to Build

0:00
AI is suddenly everywhere in the construction conversation, but
0:03
most of the industry is still trying to understand where it's
0:05
actually useful, where it's hype, and how builders can start
0:08
experimenting without needing to become AI experts.
0:11
Today's conversation explores what real AI adoption looks like
0:15
from someone actively building with it.
0:17
Hey, what's going on, everyone? Welcome to Wire to Build, the
0:20
podcast exploring how owners, operators, and builders navigate
0:24
the system's decisions and realities shaping the built
0:27
environment. I'm your host at Caravella, and
0:30
as always, I'm glad you're here because this show isn't about
0:33
chasing tools or buzzwords. It's about understanding how
0:35
work actually gets done, where it breaks down, and what it
0:38
takes to build with clarity at scale.
0:41
Season 3, we're leaning in even harder into that.
0:44
More systems thinking, more real world perspective, more honest
0:47
conversations about ownership, accountability, and the
0:49
decisions that ripple across projects long after they're
0:52
made. So whether you're an owner, a
0:54
builder, or someone living in the space in between, you're in
0:57
the right place. Let's get wired to build.
1:00
Before we dive in, I want to acknowledge our founding
1:02
sponsor, Avocado. Avocado partners with owners and
1:05
delivery teams on complex capital programs to help them
1:09
make technology actually work, aligning strategy,
1:12
implementation, and real world execution across the full
1:15
project life cycle. If you're investing in
1:17
construction technology, digital delivery, or data initiatives
1:21
and struggling to turn those investments into real outcomes,
1:23
Avocado helps teams move from intention to impact.
1:27
You can learn more about their work at avocado.com.
1:30
Today's episode starts with a moment the entire industry is
1:33
feeling right now. Artificial intelligence suddenly
1:36
showing up everywhere, and conversation and construction.
1:39
In the past year alone, we've seen AI move from something most
1:42
builders barely thought about to something people are
1:45
experimenting with daily writing proposals, analyzing documents,
1:49
exploring design options, and even helping teams make sense of
1:52
massive amounts of project information.
1:55
But like the wave of technology and construction, the real
1:57
question isn't whether this will exist, as whether people
2:00
understand how to use it in a way that actually improves the
2:03
work. O Today's conversation is about
2:06
AI and construction. Not as hype, not as prediction,
2:10
but as something builders are already experimenting with in
2:13
real workflows. What's useful, what's noise, and
2:16
what it takes for people in the industry to start exploring
2:18
these tools without feeling like they're already behind.
2:21
To explore that, I'm joined by Marcus Turner.
2:24
Hey Marcus, glad to have you. Marcus is the founder of
2:27
Constructor and has been actively exploring how
2:31
artificial intelligence can be applied inside real construction
2:34
work clothes and beyond. What makes Marcus especially
2:37
interesting for this conversation is that he's not
2:40
approaching AI from outside looking in.
2:42
He's experimenting with it directly in the context of how
2:45
construction work actually happens and beyond.
2:47
So Marcus, before we get into AI and where things are headed, I
2:50
wanted to start with getting to know you.
2:52
We go way back to the Construction Progress Coalition
2:55
AEC Summit three years ago, I think.
2:59
And so, you know, and that was a lot of fun.
3:03
I really enjoyed getting to know you there.
3:04
We were talking about. Gamification making making it
3:08
making work fun and making it feel playful so people actually
3:12
show up and enjoy being at work. Yes, of course.
3:16
I mean, that was actually our first guest on the show was
3:18
Christina Henke, who was a part of that conversation.
3:21
So go to the very first episode if you want to listen to that.
3:24
But we also talked about like logs and ledgers and and data
3:28
strategy, which was like 3 years ago in the age of AI.
3:32
It was almost an infant back then.
3:35
So yeah, let's hear your story because I know a bit of it, but
3:39
I want to make sure that our guests are aware of, you know,
3:42
who you are, how you got into construction and what started
3:44
this journey into a high exploration for you so.
3:49
Growing up, my grandfather actually owned a he was a home
3:54
builder. So but from ground up houses,
3:57
especially back in the day building at a bunch of places,
4:01
My dad was a person that was always a do it yourselfer.
4:04
So he finished his basement like when I was like there's pictures
4:08
of me when I was a kid, like hanging out in the basement that
4:11
in the first home he built like in a, you know, the play pan
4:15
playing with my little toys and him doing stuff.
4:17
But he's always been that way. Then obviously because he
4:20
married my mom, who for my grandfather owned a construction
4:24
company, he was a huge assistant in asset for him for being able
4:28
to do some stuff. So the built environment for me
4:32
has been like something that's always been around, but like as
4:36
far as work though, I ended up like interning at a nuclear
4:40
power plant, Braidwood, if anybody's from Illinois.
4:43
So I got to experience a little bit more of like what it was
4:49
like to be on a team. And at that time I'm date, I was
4:53
dating my now wife who worked as a architect, who actually had a
4:58
degree in architectural engineering, who was trying, who
5:01
was working in construction. So there was like a, a mix
5:04
between like, Hey, this is what we're doing at this, this
5:07
nuclear power plant as far as coordination and, and doing
5:10
things and the conversations that was happening about her as
5:12
far as the workplace, especially during that time, it was like, I
5:16
mean, I think million of those used to this.
5:18
Now there's some type of, you know, usually some work
5:22
agitation of, of trying to, how do you position yourself?
5:25
How do you figure out, how do you put your place in your
5:27
workplace? So I've been dealing for at
5:29
least 20 years now in various forms.
5:32
So I think when that was happening, we were just trying
5:37
to figure out like, OK, so how do I make myself competitive?
5:40
How do I make myself adaptable and flexible within the built
5:43
environment? So I was learning from from from
5:46
that space was it was about, it was really all about
5:49
collaboration and what she was learning, she was a part of the
5:54
CMAA and then she was interested in trying to do, but she ended
5:58
up going to a talk about Lean. So this is a company called
6:01
Skinner Construction, who basically they found out that
6:05
their company wasn't necessarily doing well, performing very well
6:08
and they transformed their whole company from adopting Lean.
6:11
So it was a really big talking discussion and we were talking
6:16
during that time and she was like, you know, I'm, I'm working
6:19
on projects, I'm working for the Chicago Building Building
6:22
Commission at that time. So it was very traditional, very
6:26
P6 scheduling trying to trying to go ahead and do that.
6:30
One of the sad parts about that was is that the, a lot of the
6:34
people were probably like my grandfather when I was
6:37
mentioning before they could do the work, But at the same time
6:40
they, we're coming into an era where you have to document your
6:43
work, right? So they, they had the, the
6:45
problem of I can do the task, but I don't, I don't have the
6:48
skill set. I don't have the thing to be
6:49
able to document. I don't have to do do the stuff.
6:52
So the, the issues that we came up with was like, how do you get
6:56
paid? Because if you don't document
6:59
your work, you don't get paid the and, and this era, right?
7:02
So, so basically from that we were navigating trying to figure
7:09
out what does that look like? How does what does good
7:11
construction like, what is good documentation and things like
7:13
that. So then we moved into this space
7:17
of realizing that lean and collaboration, which I saw in
7:25
the nuclear power place was like kind of the more of the way to
7:29
go about things. And so we were like, OK, so if
7:33
the people from what we can do is try to lead people by showing
7:40
them like a process and documentation and like along the
7:44
way, maybe that would be the way.
7:46
Then from there, there was pushback.
7:51
She now transferred to a different company.
7:53
There was pushback as far as in the space, like what some of
7:57
these ideas will look like, what would some of these things look
8:00
like as far as basically. So what we decided to do after
8:05
that was start a podcast very similar to this, right?
8:09
So we wanted to show that there are other people who are
8:14
thinking differently and other people who are doing differently
8:17
in the construction space. So I'm stopped right there.
8:20
But that's like kind of my journey of like where the
8:24
problem was and how we kind of navigate and got to where we
8:27
were. Well, I think it's a really
8:29
great story and that there was a lot of things that jotted down
8:32
as you were you were writing it because there's so much to touch
8:36
on there. And you know, I had just
8:39
recorded that was my first field note episode, which was a 5
8:43
minute short. And it was about the work not
8:45
being done until it was documented.
8:48
You know, the world, I think has changed dramatically in
8:52
construction where that documentation, the documentation
8:55
of the work being done was the building standing up and being
8:58
moved into and everybody moved in and it was, it was finished,
9:03
right. And now, you know, in the, in
9:05
the age of lawyers and making sure that everything was done
9:08
the way it was supposed to be done.
9:09
We have inspections and we have all of these documentation
9:13
layers. And ultimately, you know, it
9:15
helps us build better, which I think is the exciting thing
9:20
because we're being more intentional about what we do.
9:22
Not that the folks building before weren't intentional, but
9:27
it's capturing those moments in time that help us make sure what
9:31
was built was built the way it was intention intended to be.
9:35
One thing that, you know, technology has done to the
9:38
construction industry that I've observed is it's created all
9:41
this administrative burden, but we still are expected to build.
9:45
And so, you know, in a schedule that's already compressed, we've
9:49
added a whole lot more work. And this is where construction,
9:52
I think is what we struggled with that technology advancement
9:55
is that we're adding more and continue to add more, more
9:59
documentation, more data, more models, more whatever, more
10:02
scans, more drones, yadda, yadda.
10:05
With all of that more, how are we supposed to also go faster?
10:09
And so that was one thing that I thought really stood out from
10:12
your, your story here. But then I also kind of want to
10:15
go back all the way to the beginning where you said you're
10:18
kind of grew up in these spaces and, you know, as a as a kid
10:23
growing up on, you know, with a building family, what were
10:27
things that you were perhaps exposed to?
10:29
Cause one thing, And I think this is the thread we'll pull
10:32
on. One thing that I'm finding, at
10:34
least with my exploration of AI is having kind of like a builder
10:38
tinkerer mindset goes a really, really long way.
10:42
And so I'm curious, just to give you a thread to pull on here is,
10:46
you know, as a kid growing up in the construction, learning about
10:49
what building meant, how has that also found some similar
10:54
threads or maybe not when starting to do your first steps
10:58
with AI? So I think for me, what I
11:02
realized was, is going through the whole process for me as far
11:05
as building 1 of the things that for for seeing what my
11:09
grandfather and what my dad was doing, I felt like it felt
11:16
tedious and it felt like there could be a better way.
11:19
Not saying that like, I mean, I grew up in the era of like
11:22
technology was growing up around me at the same time, right.
11:25
So I was more of a person that was like into like the CAD space
11:30
and visualization of the space. So I think from there I was
11:35
like, there has to be a better way to to be able to do some of
11:38
these processes. But at the same time, from the
11:41
builder standpoint, that is also something I got from my
11:44
grandfather and I got from my father, which was, hey, even
11:48
when you're building and you're tinkering and you're trying to
11:50
figure this out, like what book can I read?
11:53
What can I grab inspiration from a different industry?
11:56
What tool do I need to be able to do those type of stuff?
11:58
So they're like, there's creativity in that, right?
12:01
So I use that kind of creativity and I just kind of moved it more
12:05
over to the software side because I kept looking at it and
12:09
I kept looking, or I started looking for the process side
12:11
too. Is that like there has to be a
12:14
better way, There has to be some type of coordination to be able
12:16
to go through this, right? So people go to to these
12:19
construction projects and there's like a lot of
12:22
frustration because there's not enough coordinations happening.
12:25
There's there's not enough, there's a lot of improvisation
12:29
happening. There is material that ends up
12:32
being either lost or stolen or counterfeit that could be on the
12:36
project, right? So, and I'm just talking about
12:39
like residential, which is more like a little bit more like the
12:43
wild, Wild West and you kind of make up your rules and make up
12:45
your process that you go commercial.
12:47
It's a little bit more like this, this structure, but
12:51
they're still mess that people very, very, very creatively know
12:57
how to profit off of and. The building building, you said.
13:02
Improvisation and building is a lot like jazz.
13:05
I think you, you got a lot of people and, and some of them are
13:09
more like the cat on the on the, on the chalkboard and, and some
13:13
people really know how to play their instrument.
13:15
But it's that improv is where I think the magic happens is where
13:18
buildings get built, but it's also where accidents and
13:21
mistakes get made. I know that that is funny
13:24
because in AI, and maybe this is a great opportunity to start to
13:28
kind of pivot into that is like AI does the same thing where it
13:32
starts to infer and that doesn't necessarily mean it always
13:37
starts to understand, which compared to a construction site,
13:41
it's the same exact problem almost.
13:43
Yeah, I think so. I think the thing that actually
13:47
got me into AI was, is is working at this nuclear power
13:52
plant. They, their coordination was
13:55
really good, like in order to change a light bulb in the
13:58
place, like to give you an example, and if you were in
14:01
somebody else's department or in somebody's department lane, you
14:04
could have up to 3 departments have to sit down to have a
14:07
meeting and have a coordination meeting about changing that
14:09
light bulb set light bulb just to make sure the principal is
14:11
there, right? And it's like, come on, you can
14:13
just change the light bulb, right?
14:15
But they wanted to have that practice in there because they
14:17
wanted to make sure that like no matter how big the risks, how
14:21
small the risk is or how big the risk is, you're always
14:23
approaching it the same way, right?
14:25
So, but within that I was thought like coordination was, I
14:30
looked at coordination being at the highest level, right?
14:32
But there was a statement made by one of the executives.
14:36
This is before AI. Well, I would say before AI.
14:39
He was like, we have all this data, we document everything and
14:42
coordination. Yeah, because I worked in the
14:45
the Phil's engineering maintenance department.
14:47
So I was work hands on with you. I was the line before the, you
14:52
had the maintenance technicians that did the work.
14:54
Then you had the people who were helping coordinate the packages
14:57
and doing the work. So I was like, right, I was in
14:59
that space. So what they've noticed was, is
15:04
like, hey, we can't Google all this information that we are
15:07
actually collecting or, or looking at all this information.
15:10
We don't have a tool to be able to do that.
15:12
So that's kind of set with me for, yeah, maybe about maybe 15
15:19
or 18 years. And that's always kind of the
15:21
idea is always kind of set with me.
15:23
And so when AI came across, when the AI came about, the first
15:29
thing I said to myself was, oh, this is going to be the version
15:35
of Google that's going to be out there.
15:38
So there were two things I ended up doing.
15:41
So I logged in and grabbed the chat TPT account.
15:45
So chat 3 point 3.5. I went in and said, hey, look at
15:50
this document getting tell me what the information that you
15:53
see in this document is. So I copied to paste document
15:56
in. And then I said, oh, OK, can you
15:58
explain it to me as a 5 year old?
16:02
Can you explain it to me as a 10 year old?
16:04
Can you explain this to me? If I was give me another analogy
16:08
like say if I was cooking or something like that and I was
16:10
just messing around with it, trying to just feel like what I
16:14
can gain from it. And the thing that I got from it
16:16
was there were two things that I got from it.
16:20
Now everyone tries to document the work in the best way they
16:24
possibly can from their perspective, right?
16:27
And this allows you to create a tool so somebody can ask a
16:33
question in a different way. And the output could be in the
16:37
way that the person needs to receive the information.
16:40
So I mean, that was like a really big unlock because
16:44
there's plenty of times where I see like documents and stuff
16:46
like that. I'm like, man, this is like over
16:48
my head, right? Or, and you know that the
16:53
person's trying to the person who's done it, they probably
16:55
done it with the, the best care they possibly could in the
16:58
world, but I still don't understand it.
17:02
Now I can have something to be able to do that right now.
17:05
I can have a conversation to somebody and actually talk to
17:07
them at their level because I have understanding or I have a,
17:10
a, a vehicle. In this case I would say to do
17:15
some type of knowledge translation.
17:16
Yeah. And that's too like in
17:18
translating it back to construction is one of the
17:20
biggest things is like an engineer is going to understand
17:24
a system in a very different way than even the person installing
17:27
that system is going to in the same, same but different.
17:32
It's just very interesting to think about, you know, how
17:36
people work together and where how important communication and
17:39
collaboration is in construction and where, you know, there's
17:45
that that age-old picture that goes on the Internet where
17:48
there's two guys standing and they're looking at the either A9
17:51
or A6 and they have a fundamental disagreement just
17:54
because of where they're standing.
17:55
But it's the same object at the end of the day.
18:00
And this is where I think, you know, going back to inference,
18:04
even, you know, having without having understanding, you leave
18:09
people to try to understand. And that inference is where
18:13
something bad can happen. And this is likely where in the
18:17
nuclear facility where it might feel pedantic to have, you know,
18:21
three guys sit in a room to figure out how to switch a light
18:23
bulb. But it also might make a lot of
18:27
sense because, you know, you might have a maintenance guy go
18:30
out and he's now got a ladder and he goes, he's switching the
18:33
light bulb. But there's something that has
18:34
to happen. Maybe somebody has to walk down
18:36
that hallway at a specific time, do a specific other thing that's
18:40
essential for the operation of a facility and he's in the way.
18:44
And that could be that could be your problem.
18:46
There is things that seem obvious without context.
18:52
And this I think where you very early on realized like you
18:57
didn't ask AI to give you a definition before you uploaded a
19:00
document, which I think a lot of people do.
19:03
I don't know if you've heard it. Like I've heard people go on,
19:05
it's like I asked AIAA code question and it got it
19:09
completely wrong. And I'm like, well, did it have
19:12
the code uploaded into it like where you're just using LLM?
19:18
And so if you're new to AII think lesson #1 for the, for our
19:23
listeners is like, make sure that you give the appropriate
19:27
context to an AI before you start to query it.
19:31
Would you would you agree like that's fundamentals or?
19:34
I would, I would say that was probably one of the biggest
19:36
things. You know, there's a phrase now
19:38
that's called context engineering.
19:39
And what is really just saying is that you need to give it the
19:43
right context, right guardrails, right constraints.
19:46
Even trying to tell it what not to do would actually steer it
19:50
into more of a direction that you may want it to go into.
19:54
And then, you know, listing what your goal is, right?
19:58
You'd be like, hey, this is my constraints, blah, blah, blah,
20:00
blah. Don't do these things.
20:01
And here's my goal. And here's the context that I
20:03
think you need for this information.
20:06
And then you can say, hey, you know.
20:09
Here's some here's another piece of my objective and then it'll
20:12
spit out what you need right. So I think being able to know
20:16
that those are kind of one of my best one of my if I give a
20:20
presentation, I give a talk to people who who don't or first
20:23
time experience in AI or think they didn't they or trying to
20:26
understand AI that I the habit is is that we like to think that
20:32
it's supposed to function like Google and it's not a search
20:35
engine. Now you can tie it to a search
20:37
engine, but it has a fixed training for a fixed amount of
20:41
time and a fixed moment of time in history, right?
20:44
So every training date is set for like, for example, like the
20:50
newer models that are coming out are usually set for are kind of
20:54
some of them are set for February and January that and
21:01
maybe sometimes then, you know, a few months will go by and then
21:05
they have a new another training cycle.
21:07
And then it's more up to date with the new information.
21:10
But when the new model comes out, everyone's, you know,
21:14
tinkering with it, doing different things with it and
21:16
stuff like that. So like the information, what
21:19
the model knows, it's not what the information what society is
21:22
now using the model for, right? So, you know, I can I can give
21:28
an example of a tool that we can talk about later on.
21:31
But like clock code was a example of a tool that people
21:35
thought was just for coders. And then they realized that
21:39
people were doing it for non coding tasks, asking it to do
21:43
different things and being able to unlock different things that
21:45
they, the developers themselves didn't even know was even
21:49
possible. So this really reached this kind
21:52
of change how knowledge work is looked over the past, probably
21:57
about a year now. Now it's had this huge unlock
22:00
that is different than what it looks like, how we used what the
22:04
assumption was back then. And it's interesting to think
22:08
about like that impact, especially in the world of like
22:10
image generation and video creation, how far fast that
22:14
that's come is like, I think as soon as the new the new training
22:19
model comes out, the whole world jumps and starts to pressure
22:22
test these things. And you get those like edge
22:25
cases, those fringe cases, those completely upside down.
22:28
I never thought we'd use it that way.
22:30
Case in some ways you get like these really cool aha moments
22:34
with cloud code, which I know we'll get into because it's like
22:37
my next frontier of you know now I can be making something or I
22:43
can be organizing information in other ways and keeping kind of
22:48
with the early level like when I've I jumped in I was using
22:51
ChatGPT. I used it like a search engine.
22:54
I was impressed that my dad used it like a search engine.
22:56
I was because he was like, hey, he adopted new technology.
22:59
That's that is a signal I've. Still kind of get my dad.
23:03
Yeah, but it was like always this kind of query back and
23:07
forth. And then I realized like, OK, I
23:09
can start building like knowledge sets and information
23:13
sets and, you know, have it reference different things and,
23:18
you know, become a creative partner, which I thought was so,
23:22
so, so fascinating. And then we get down in an
23:24
agentic happens. And I, I would love for you to
23:28
help people understand maybe like the difference between like
23:31
probably LLM and and the basic search engine relationship.
23:36
And then what agentic starts to mean in the scope of AI is.
23:42
So there's a lot of definitions that are out there about what an
23:47
agent is or what what agentic framework or what, what, what
23:50
the whole so-called agent looks like.
23:53
I've summarized it in a very simple way.
23:56
I was just saying it's an LLM call and you give it a tool and
24:01
you put it in a loop. That's it.
24:03
Search can like, let's go back and talk about what I said about
24:07
the using as a search engine. Search can be used as a tool.
24:11
So it leverages its previous knowledge or its training
24:14
knowledge, then it gets the additional information out and
24:18
then it goes back and then gives you the information that is an
24:21
energetic pattern that is information that you can get.
24:25
And then you can, you can even loop, like for example, you can
24:29
loop it even further, right? So there's researchers that are
24:33
out there right now that are finding that if they let a task
24:40
run long enough, they can potentially find scientific
24:44
breakthroughs. So there's one that just
24:46
recently passed for physics that for a scientific breakthrough
24:50
that Chet GBT discovered. And so the question is, is that
24:57
if we let it run on a long running task, depending on how
25:02
hard and if you wanted to solve a hard problem, what other kind
25:06
of breakthroughs can we get in this space?
25:09
And that's just only about search and research and it
25:12
generating the information, right?
25:14
So now the question is, is what else can we build, right?
25:19
So now we can go into like, maybe like for example, we
25:26
listen, we can move into the like the coding frameworks
25:28
because that's what things were made for at the beginning was
25:31
like, hey, let's help the engineers because we are
25:35
engineers, you know, it's like forest by us, right?
25:38
We're going to make the product for us.
25:40
So we can go ahead and be able to use the product better,
25:42
right? So from there they were like,
25:44
OK, so why can't we use it with code, right?
25:48
So why can't we build this agenda framework for code?
25:50
Because code versus natural language code is very limited as
25:54
far as like characters, things you can do, scope and things
25:57
like that. So they thought that would be
25:59
the easiest framework. Crazy enough, they end up being
26:02
the big unlock for all the other language work because we
26:05
basically use code to do tasks. I used, you know, we weren't,
26:11
you're using Riverside FM right now, right?
26:14
That was built by somebody using code to be able to do that.
26:19
Well, one day we may not need to in the in gentic world, we may
26:26
not even need to open up app applications anymore.
26:30
Just imagine. And we're seeing it now play out
26:33
within Co worker. If those who know what Co work
26:35
is, we're seeing it play out in Co work.
26:38
Co work has access to all these little different information, be
26:41
able to pull different things together and stuff like that.
26:44
And it's in just one hub because the agents don't care where the
26:48
information comes from. All I need to do is be able to
26:51
access it via API, via CLI, which is now becoming a thing,
26:57
which is command line. Those who remember log into the
27:00
DOS and playing and trying to play your favorite video game as
27:02
a little kid, you know, we're going back to that world in a
27:05
way, but we're just telling the agent to go ahead and do it.
27:09
Or, you know, who watched the sci-fi films and remember the
27:12
computer that you just typed a random question into the
27:15
terminal and the computer start talking back to you and you now
27:18
you and and and showing some type of personality.
27:22
That world is starting to become a thing in our reality now.
27:26
It's yeah, that was, it's, I had a insight two years ago when
27:30
this is all starting to really catch fire.
27:32
And it's one of the we do a lot of custom had done a lot of
27:36
custom integrations at the at Avocado where I was working.
27:39
And I was like, if we start giving agents access to these
27:43
solutions, we don't need to build custom integration
27:46
frameworks anymore. The AI is just going to go, it's
27:50
going to log in as if it were a user, It's going to go look
27:53
under 1000 rocks and it's going to come back with the
27:55
information that it needs. And this is going to rapidly
27:59
change an entire segment of industry because the need for
28:04
inter, the way we think about interoperability is going to
28:08
vastly change because it is going to be able to just have
28:13
access to certain apps and look at certain files and look at
28:17
different things. And there's this thing that I
28:21
really get excited about when it comes to AI, which is like the
28:24
democratization effect, which is that anyone individual now has
28:29
the capacity to code or create a video or create a song.
28:34
And that is democratizing creativity and curiosity in ways
28:42
that, like, you used to have to be good at something in order to
28:45
be able to do something. And now you can have access to
28:48
applications or even create your own applications to be good at
28:52
something. Yeah.
28:54
I can give an example of a person that I just helped.
28:58
I I'm supposed to even do this video of him, but I haven't, I
29:02
haven't done it yet, but we talked about doing it before.
29:05
This podcast was funny. We just didn't have time.
29:07
So his name is Jeff Riley. He he is a person that is in the
29:13
home construction space and he has a tool that he was trying to
29:18
understand that the he wanted to basically test Project Health on
29:24
a construction site. He came to me with this idea.
29:29
I didn't know if he was willing to like, I guess see it through.
29:33
What did he know? I didn't know what he knew about
29:36
AI or anythings like that. So he said I have a chat GPD
29:40
account. I was like, oh, then I was like,
29:43
well, are you using it or how well how deep are you about to
29:48
do this and things like that. And he was like, I don't have no
29:51
knowledge, no nothing know anything about this.
29:53
I was like, OK, so the place that I usually suggest people to
29:57
go to is like Google Gemini, because Google Gemini.
30:01
Well, Google is, you know, they're a giant that they are.
30:03
They have a lot of free tools to be able to visualize and be able
30:06
to create stuff. So they had Google has a product
30:08
called AI Studio, which they just revamped yesterday, which
30:12
is as as a vis according to 90s. So, and one of the things that
30:20
one of the things that he did was we took his idea was like,
30:25
we'll just test it out to see how it looks.
30:27
And he got I would say about 80% of the way there by using the
30:32
tool. I coached him all the way
30:34
through. It didn't have to do anything or
30:35
anything like that. We were trying to, this is
30:38
before they did the re rebrand. It's a lot easier now, but we
30:42
took the code down and then I said, OK, you have a $20.00 a
30:47
month jet GBT description. Open AI has their version of a
30:51
coding tool called Codex. So we took that code, took that
30:55
in about an hour and a half. He had a working application
31:00
that he launched, showed and now demoed.
31:03
And now he's working on his second application to try to see
31:07
if to work on a different problem that he has or different
31:10
thing that he's inspired by. And one of the things that
31:14
surprised me about this conversation or, or going
31:16
through this whole process was is that if this is somebody who
31:21
has never gone through this experience before.
31:23
And all I had to do was like coach and hand hold a little
31:26
bit. And he was able to get to where
31:28
he, he, he was able to execute an idea from there.
31:31
You know, what does that what, what is the world of software
31:34
going to look like that so many more people can now execute
31:38
these different things because everybody's going to have their
31:42
own little tailored outfit now to fit their special use case.
31:46
Well, and I think as we're really getting close to the the
31:49
hype side of this, I, I would love to think about what is
31:52
maybe some of the caution or or not like what are things that
31:57
people should be considering? If you heard you heard these
32:01
stories. Now you're like, I'm going to go
32:02
build myself. Whatever.
32:03
I'm going to go build because I can build anything.
32:06
And that's the excitement we want you to leave this episode
32:08
with. What are things that people
32:10
should be thinking about as they're building the the things
32:13
that they might not know? And one thing I always ask the
32:17
AI even is like, what's the thing that I don't know?
32:19
And it usually is pretty good at it.
32:21
I have some ideas. I'm curious what your your your
32:24
thoughts are. Your cautions are about, you
32:26
know how to go about building, where to be careful and more and
32:31
potentially thoughtful is maybe the more the right word when you
32:36
start to build. So I think one of the things
32:38
when you start to build, you have to worry about security.
32:41
Security's number one. So there's a lot of things that
32:44
have came out that people have vibe coded that have gotten
32:47
famous. There's a guy who, I don't
32:49
remember the vacuum cleaner company, but he wanted to put
32:53
his control, his play PlayStation controller to his
32:56
vacuum. And so he wanted to be able to
32:59
use that, which is cool, right, and make my vacuum move left or
33:02
right or whatever. Because he has Bluetooth, right?
33:05
He found out that he was able to do it and then he went on the
33:09
website to, you know, view his vacuum cleaner to see if it was
33:13
registering. Well, he controlled the entire
33:15
fleet of vacuum cleaners. I don't know what was going on
33:20
in the other houses. And this is a company that's
33:22
been around for a few years. And it's not, this is not a
33:26
company that like probably took care of their security and
33:29
really concerned about it at one point.
33:32
This is just a guy who was just hacking his controller and
33:36
vacuum together and realize he's Cod and Co fleet.
33:39
And I think those are the things that could be security concerns
33:44
when you start building these websites, right?
33:46
So like being able to make sure that you are keeping your
33:51
clients data private when they actually sign into those
33:54
websites, especially when it goes viral, because when it goes
33:57
viral, everybody's going to be trying every sort of thing to be
34:00
able to get access to the information.
34:03
So they can either resell it or do other things with it.
34:08
You know, I think that's it's a predatory practice, but I do
34:10
realize it's a really good practice.
34:12
I actually have there's even cybersecurity companies that
34:14
build AI agents that they hear any topic that's in a viral,
34:18
they will actually go and try to ethically hack it and then try
34:21
to offer this their their services to these companies,
34:24
which is a smart thing to do. But at the same time, now you
34:28
have good actors who are chasing capitalism and you have bad
34:32
actors that are trying to chase and go after it too.
34:35
So you're going to get hit in different ways and things like
34:38
that. So I think just security would
34:40
be the number one thing. Both the bigger houses are
34:44
thinking about that and, and, and looking into that.
34:47
So Google Claude and I mean anthropic and open AI, they have
34:53
their, they actually have solutions that are out and
34:55
available too, so that you can like use those solutions.
34:58
There's other, there's other firms that we haven't talked
35:02
about it, but they have these things called skills, which is
35:05
you just basically type in, Hey, use this skill.
35:09
It will look at your code and actually look at some of the
35:12
vulnerabilities that are there. And there's some other companies
35:15
that have it from a, from a low cheap cost that you can actually
35:19
look at your write your code and be like, hey, OK, can you review
35:21
this code to see if there's any like issues?
35:24
I mean, because everybody's going to miss stuff anyways.
35:26
And don't let nobody tell you that just because people, humans
35:28
made the code doesn't mean it doesn't have security issues in
35:31
it. Because that's, that's, that's a
35:32
lot. So there's always something
35:36
people will always miss things. And also, you know, these
35:39
systems are also made by humans. So some of our mistakes are in
35:42
them. So security is definitely one
35:45
that I would say when I personally experienced too, just
35:48
like the hallucinations, hallucination or even failure,
35:51
persistence. And so something I've learned
35:54
generally when using AI is like document and re document and and
35:58
keep as many tight documents as possible.
36:01
It's in the in the gaming industry, which I know Marcus
36:05
and I are big fans of, is like having a saved state and being
36:09
able to revert back is always a good thing.
36:12
And so that that would be another thing I deeply recommend
36:16
is like as you start to play around and explore any concept
36:20
or idea. I'm a big fan of Notion.
36:23
I know in code you can actually just create yourself
36:25
repositories and markdowns, but creating reference documents not
36:30
only for yourself but for the AI to reference has been a huge
36:34
unlock for me in the space when I need to be accurate and I need
36:39
to be precise about what I'm doing.
36:42
Yeah. And I would say another thing to
36:43
add on to that is they have a. So we talked about context
36:47
engineering and they so there's two things that people call it
36:51
context rot or the dumb mode. Dumb mode, which means if you
36:55
have long contacts that you're having for these conversations,
36:58
try not to overextend your conversations.
37:01
Get your task done wherever that task is and then move it to
37:05
another window. So you're still keeping the
37:08
model paying attention. It's just just like how we get
37:12
overloaded with contacts or something like that or, or too
37:15
much information. There's just a performance job.
37:18
And I know that these models are coming out with these 1,000,000
37:21
token windows right now, which is, you know, it was crazy
37:25
because when we're looking, if you look back at it, it was 16 K
37:29
32K and 128K was the, that was just three years ago, which was
37:35
considered the context window, even though there was research
37:38
that pointed out that the 1,000,000 context window was
37:41
coming. So we just, you just didn't
37:44
think that was going to happen in three years that we're going
37:46
to have this 1,000,000 contacts window.
37:48
But the, the performance actually does still degrade
37:52
within those 1,000,000 contacts windows.
37:53
So if you have a lot of the information that you want to
37:56
dump in, it's still good, but it's still not going to perform
37:59
as well or be as performant as maybe a system or a situation
38:04
where you have small, this is going, I'm going to go into a
38:07
genic framework a little bit, but you want to have small sub
38:10
agents go out and do the research and then report back to
38:13
the bigger agent. So everybody has their context
38:16
windows pretty lean and you'll be able to do what you want to
38:19
be able to do. But if you're using ChatGPT or
38:22
Microsoft product, just try not to have your conversations too
38:25
long. And then as you finish the task,
38:27
just pass it off to another context window and then keep
38:30
going. So before we go into kind of the
38:33
adoption gap and why people may hesitate, I would curious about
38:38
this question that actually came through an AI prompt.
38:42
But would you find that there's, I feel like there's a probably a
38:47
shared sentiment here, but are some people using AI as a
38:50
shortcut instead of understanding the work?
38:53
And I would add to the end of this of how truly important is
38:56
it to actually understand what you're trying to do when using
39:01
AI? I think so.
39:06
I have young kids, right? So like, I think my analogy is,
39:10
is that like, you know, kids don't necessarily they see a
39:14
concept and they can see other people walking, but they don't
39:16
really know how to walk. And we're trying to do that with
39:21
AI, with some of the activities and things that we're going to
39:23
do. Those who excel most are going
39:25
to be the ones who have domain knowledge of the topic.
39:28
So like a coder who understands what a software stack and how to
39:31
put a software stack together and all those little different
39:34
information right there that if they embrace this, it's a huge
39:38
unlock for them because now they can be twice their coder.
39:42
I've had friends that are coders that they, they talk about.
39:47
I finished all my pool requests within an hour now and now
39:50
they're doing other things at work.
39:52
Some companies, Stripe for example, put out a report.
39:57
They said that they're able to finish 1000 PRS in a week now.
40:02
So they're and which is really impressive.
40:05
But if you still have domain knowledge about something like,
40:07
if you're like, not necessarily a coder or anything like that,
40:11
like you're going to have an advantage because you're going
40:14
to know the work better than. Even the, you know, I gave an
40:17
example of the developer, but you're going to know the work
40:19
and you're going to know the process better than they do.
40:22
And most of your time talking back from them is like a
40:24
translation back and forth like I was talking about before,
40:27
which is, oh, do you mean this or do you mean this or do you
40:29
mean this? Right?
40:31
I used to work as a photographer and one of the things I used to
40:34
tell the models or I'll tell like a makeup artist especially,
40:37
or hairstylist is show me what you're talking about.
40:41
And I think AI has given us the point that we can now show our
40:46
work, show our ideas, show our things.
40:49
And then somebody else can come in or, or another person or
40:52
another partner could come in, could be somebody, a second, or
40:55
could be somebody that's not and be like, oh, OK, have you
40:57
thought about this? Have you thought about this?
40:58
And then the process gets improved and gets refined over
41:02
and over and over again. And you know, and AI for me,
41:07
I've, I think of it as it's called, it's solved the Blake
41:10
sheet problem to me because sometimes I get nervous about
41:15
like, oh, OK, you need to write this report or you need to write
41:17
this information. It's like, OK, so where do I
41:19
start? I don't have to do that with AI
41:22
anymore. Because ultimately, I think
41:25
this, this is really good. The way to tie it to is like
41:27
having that understanding in that context, you know, what
41:30
good looks like. So if the AI gives you a bad
41:33
idea, like it's not, Yeah, like we were doing a website refresh
41:38
and like it was telling me, oh, well, you know, your, your
41:41
viewer, your, your visitors are going down.
41:43
And I'm like, yeah, but I'm not getting any action from the
41:45
visitors anyway. So I actually don't care about
41:47
this, this thing whatsoever. So forget about that data point.
41:52
And so it was really interesting to think about like how
41:55
important context is. But then this probably goes
41:57
right into the second part of the problem, which is where, you
42:00
know, folks fear we're getting rid of entry level jobs.
42:04
And so if we get rid of entry level jobs, which is where
42:06
context is curated, how do we how do we as people build a
42:13
right approach of building context in new frontiers?
42:18
And how is actually AI probably part of being able to help to
42:22
build that context? One of the things that I think
42:26
it's going to be a really good thing to be able to unlock is,
42:29
is the for the older generation to be able to speak to the
42:32
younger generation. So a new person comes in, is
42:36
trying to learn from the older person who's been there for, I'm
42:39
going to say I'm not 20 plus years.
42:40
I'm going to say 30 plus years, right?
42:43
There's a, there's things that mistakes have happened over the
42:47
years, things that like I'm, I'm thinking there's just
42:50
institutional knowledge that's there.
42:52
However, because of the young person doesn't know how to ask
42:55
the right question, it doesn't unlock for the older person.
42:58
But Oh yeah, I forgot about that, blah, blah, blah, blah,
43:01
blah. They'll tell you the whole
43:02
history and give you a story about it.
43:04
You'd be like, yo, like you could have told me that.
43:06
And so I don't have to make that mistake back there, right?
43:08
So. Sometimes it's funny to watch
43:11
people make the mistake. Yeah, He's like, hey, I remember
43:13
that mistake. Now you're not going to do it
43:15
again, right? You just had to go through it,
43:17
right? You know, it's like, like some
43:19
things in life, you know? Yeah.
43:21
So, so I think one of the things that that this allows or unlocks
43:29
for people is the idea of OK, older person, I can now transfer
43:35
my knowledge to this person and I don't need to necessarily sit
43:38
there all day, right? So like I could be interviewed
43:41
by the younger person too, and I could just take that transcript
43:44
and start cataloging these different transcripts of this
43:48
information and that can be also into institutional knowledge.
43:51
Vice versa, on the other side, the younger person can now be
43:57
more capable and almost start doing what the, I would say the
44:01
mid level person could do because you can start giving
44:03
them a little bit more riskier tasks and things like that.
44:06
Like, oh, I would have gave this to my middle level.
44:08
And while I went through that to my more experienced, not saying
44:12
senior, senior PM, but maybe I would have gave us to my mid
44:14
level PM or something like that. And you can be able to give
44:17
those tasks to be able to do whatever those things are,
44:19
right? So I think sometimes that
44:21
there's going to be this enabling that's good gets
44:23
unlocked. In addition to that, the younger
44:25
person may not necessarily be burdened by the past of what has
44:32
happened. And so you have more fresh ideas
44:34
that happen. But there's a counter to this.
44:36
And I will say the counter, the counter to this is basically I
44:40
don't need to hire those young people to do those small simple
44:42
tasks anymore. And I'm just going to make the
44:44
people on my current team more capable by enable them to be
44:48
able to use AI. So some some people are going
44:51
through the some corporations are going through this
44:53
experience of nothing written by hand, everything done by AI.
44:57
And that means all departments or all departments must ship
45:00
something or create something and show something from AI.
45:03
And they will have set up like weekly meetings.
45:05
So everybody can kind of cross learn and cross learn from from
45:09
the whole situation. Tell me what you've done with
45:11
AI. Tell me what you've done with
45:12
AI. Tell me what you've done with AI
45:13
to be able to get those things, which I think are really forward
45:16
thinking companies because they're thinking about this
45:18
future that's going to that's this out right now.
45:20
And they're gaining more capability from their employees
45:23
and they're sharing ideas and getting improving and working
45:26
better. But at the same time, some of
45:28
them are like, Hey, maybe I don't need I can fire all these
45:31
people or my my next hiring, right, I'll have to hire all
45:34
these people either. Yeah.
45:35
And that I think is, well, that's one of the complexities
45:37
is as a large institution, I do see you're out there risk of
45:42
kind of being not top heavy, but just heavy now because people
45:46
can do so much more. And so large corporations are
45:49
going to come to the fire. Like Accenture was like getting,
45:52
you know, under fire for all. Like, hey, we're moving AI and
45:55
all these people are getting laid off.
45:57
And then in a small company, it's like, hey, we're a three
46:00
person team of AI and we don't need to become a fifty person
46:03
team because we can do health of these things and that becomes a
46:08
emotion of success. But let's then use that as I
46:12
think a little bit of maybe hype and excitement for the people
46:15
who maybe are finding their way or want to start their own
46:18
thing. And this is the adoption gap of
46:21
like, start using it now is that, you know, we're creating a
46:25
world where, in my opinion, you know, small businesses are the
46:29
only businesses moving forward because they're going to be able
46:33
to reach further and farther and be able to collaborate with
46:36
other small businesses to cover their bases.
46:39
Not that there's not a space for big businesses either, but I do
46:42
think that the competitive angle between large and small business
46:47
is going to start to compress really meaningfully with AI
46:51
advancement. And so, you know, with that, you
46:54
know, if you're this younger person who maybe doesn't have
46:57
this context or even somebody who's never used AI before, how
47:01
do you start to adopt it today? People are feeling like they're
47:07
behind. How do you get over the fear of
47:09
getting it wrong? How important is experimenting
47:12
and and what does it mean to get started?
47:15
I think it's actually so with the tools now and some of the
47:19
people that I know that are stepping into the journey and
47:21
starting to use some of these more agentic harnesses and more
47:24
of these agentic tools, I think now is actually better than when
47:30
I was starting. I was copying and pasting back
47:32
and forth from ChatGPT. So like my feedback loop was
47:35
like kind of it's like, oh, I got to open up a window and then
47:38
close a window, then put this in.
47:41
Did it compile? Did it work?
47:42
Oh, OK, I got this error. Now take this error, copy and
47:45
paste it, put it back. OK, I can do all that from one
47:48
window now. Or if it finds an error, it will
47:51
just fix itself from it. And, and it doesn't mean that
47:55
I'm not even talking just about code.
47:57
It could just be I've been just trying to research to be able to
47:59
build a document, right? It'll say, oh, I came into this
48:02
error, blah, blah, blah, blah. This is why I'm going to go
48:04
ahead and do instead. OK, you've now fixed yourself
48:08
and recorrected. I think it's actually a lot
48:10
easier now and the it's more intuitive as far as what people
48:15
would expect or, or what would look for from, from AI.
48:19
So now it looks like, hey, I got this thoughtful partner or this
48:22
partner that's actually working with me.
48:24
And you know, I'm going to use a Star Trek if me.
48:27
So make it so #1. So you, so you're basically
48:31
telling me, here's my idea. Make it so.
48:34
So those who are Star Trek fans hopefully appreciate that.
48:38
Ellen, it's, you know, actually hearing you say that back to me,
48:42
it is, it's creating this compression and like the
48:44
learning, the, the learning is now faster and the AI is more
48:48
informed. So the curve of learning has
48:51
gotten shorter because it is this using it today as opposed
48:54
to using it even months ago. So, and sometimes there's model,
48:58
model updates that sometimes I don't like as much as I like, of
49:01
course, older adults, but the, the learning curve, the learning
49:06
has compressed as much as the advancement curve is
49:09
compressing. And so it goes back to that
49:12
democratization idea like it is actually now is the time to get
49:16
into it. It's never really too late
49:18
because it's the curve is getting easier.
49:20
But don't wait either. So really looking forward now.
49:25
I wanted to, you know, we talked a lot about like early use and
49:28
some of the lessons we learned, but you are doing some
49:31
incredibly cool stuff and that is 100% what I wanted to make
49:34
sure we had some time to talk about today.
49:37
So how is, you know, what are, what are some of the things that
49:40
you're working on now? It's a little bit more advanced
49:42
that you're up and sharing about and like, where do you see the
49:44
the future of builders in the AI world?
49:47
I think a lot of people that you're going to see that are
49:51
maybe either superintendents or more senior tradespeople, you're
49:57
going to see them start building their own solutions.
50:00
You're going to see that a lot. There's a example of, I know I
50:05
gave my own example, but there's an example that's on on X that
50:10
went viral. That's a 10 minute video or 13
50:13
minute video of a contractor who was able to build a solution
50:20
using claw code, didn't know anything, copy and paste that
50:24
everything in. If he didn't know something, he
50:26
would screenshot it and just send it in to Claude and ask,
50:29
hey, what do I do next? And he was able to build an app.
50:33
He just talked about, you know, after it went viral, he has like
50:36
100 new users that are looking to, to join and use his
50:39
platform. And I saw a lot of inspiration
50:41
from a lot of the other different builders.
50:43
And I was like, hey, if he could do it, I can do it too.
50:46
So you're going to see a lot more of that starting to grow
50:49
and starting to happen. I've I've seen it even from the
50:51
FM side of, of people who are coming in and saying, Hey, I was
50:55
a Carpenter and now I'm a a software entrepreneur.
50:59
Like how did this happen? I mean, but I think also just
51:03
internally, you just may be able to build your own kind of tool.
51:06
You know, your Excel's Wizards that are at the companies,
51:09
right? They're going to be the people
51:10
that are going to be probably the most prime for some of
51:14
those, some of this stuff or your Smartsheet Wizards or
51:16
whatever the tool is that you got to, you guys kind of kind of
51:19
used to get these people to do stuff.
51:22
They're gonna be the people that are like really, really, if
51:25
they're given the tools, they're gonna be the people that are
51:28
gonna really XLS far as I'm coming up with some creative
51:31
ideas. Trying to remember the second
51:32
part of your question right now. What was the second part?
51:35
I was just loving where you were going.
51:36
I was actually gonna kind of follow that up.
51:39
OK, follow it up. Yeah.
51:41
You know, there's a lot of AI tools that are out there and
51:44
you're talking about people who are building things with cloud
51:45
code and you're building, you know, and you talked about
51:48
ChatGPT. What does your AI stack look
51:52
like and what do you like? Because I'm finding personally,
51:55
too, that certain AIS are better at certain things.
51:58
And you talked like, think Google too, about image
52:00
generation. What?
52:01
What does your AI stack look like and how do you get them to
52:05
kind of work together or against each other?
52:08
I've, I've tried to say as agnostic as possible, I do own
52:13
the $200 Chechi B subscription, but I also own the mid level of
52:20
whatever the cloud subscription is.
52:23
Even though the cloud subscription in the mid levels,
52:25
they, they really want you to just get the same level
52:28
subscription that Open AI has. But I still use it and I have
52:35
some, I have a lower level of subscription in Gemini and
52:40
there's different reasons for different tools.
52:42
So like for example, if I'm going to send, the cool thing
52:46
about Gemini is I could send a video, I can send an image and I
52:49
can send a text, I can send APDF.
52:52
And Gemini will do a pretty good job with all those things.
52:54
The reason why I said PDF is because PDF can be a mixed file.
52:57
It can have graphs, tables, pictures, It can have a whole
53:00
bunch of stuff in it. I, I would say, I would suggest
53:03
that if people can create markdown files or text files,
53:07
LLMS love those the best. And you'll be able to get a lot
53:10
of contacts out of those things. If you can't create a CSV file
53:15
and maybe the IT will do a good job would be able to do that.
53:17
But Joe, just those things, text files are usually the best for
53:21
LLMS. So, but I think Gemini is really good for that.
53:25
If I want to use Claude, I always call Claude the more
53:28
ambitious person. So if I have a kind of a vague
53:31
idea or something like that or I have a vague execution of an
53:35
idea, Claude would be actually really good for that.
53:37
So I would or Claude's actually really good.
53:40
Claude and Gemini if you're built, if you're a builder and
53:42
you want to build apps or even presentations.
53:44
Anything that has anything to do with like front end
53:47
presentation, viewing, look, taste wise, I prefer Gemini or
53:51
Claude over ChatGPT or open AI in general.
53:56
Yeah, Claude's not afraid to tell you you have a bad.
53:58
ID yeah, yeah. But the nice thing about South
54:03
Opening eye, what I like about it is it does a lot of like good
54:06
checking. So like if Claude does something
54:09
or something like that, like it will it will give you that that
54:12
dynamic of like, hey, did you look at this?
54:15
Did you check this? Did you check this or whatever?
54:17
I'll do that with open AI. And then in addition to that, if
54:20
I have a lot of context and I want to synthesize, I'm usually
54:23
throwing that into open AI first.
54:25
So I'm like, oh, OK, I have a lot of ideas.
54:28
I end up being an ideas guy. I built about, oh, I know where
54:33
you're going with it. So I've built like various
54:35
different apps, sort of various different apps for myself.
54:37
As far as me going about like the early days of the process
54:41
into where I am now, where I was in early days, I give an
54:43
example, I built maybe 4 apps that are like internal use of me
54:47
being able to do stuff. Today I'm at probably maybe 9 or
54:54
8 and I'm done it in a smaller window.
54:58
So within a 2 month window I'm kind of at that .3 that I'm kind
55:04
of in a part of where I'm like, OK, I think I can release this.
55:07
So now I'm at this. So I'm at this point now where I
55:10
think when I when I look at my tech stack where I look at like
55:14
the different models alone, those are the things that I'm
55:16
looking at. I'm just like, OK, Also I prefer
55:20
the plans that come out of Open AI versus the plans that come
55:23
out of Anthropic. So I was like, Hey, look at this
55:27
plan or and then I would give it to anthropic and I'd be like,
55:30
Hey, take your little ambitious self and then improve upon the
55:33
plan. So that's like another thing
55:36
that I like about the anthropic because anthropic will they
55:40
it's, it's good because it thinks about the things that you
55:43
don't necessarily consider. And sometimes it's half hazard,
55:46
which is OK, but at the same time it's like, oh, OK, this is
55:50
kind of where I want, want things to be so.
55:54
Yeah, it's funny. It's pretty similar to like how
55:56
I how I felt about it too is like Claude is really good at
56:02
to. I also think Claude is really
56:03
good at like iterating and executing on things in ways that
56:07
the other solutions aren't. But I usually like Chachi BT
56:11
sandwich it where I'll I'll open AI, give me some give me a solid
56:15
plan and strategy and then I will execute the heck out of
56:19
that thing using Claude. I would not execute it in Open
56:22
AI because Open AI is going to kind of pontificate and use,
56:26
which is what I love about it. But then Claude is just going to
56:29
go in and hammer through that thing.
56:31
It's almost like a contractor versus an arcs.
56:34
And then after they're done building, I'm going to go have
56:37
the architect. Like, is this what you asked
56:38
for? Exactly.
56:40
So it's really kind of funny way to hear it back and think about
56:44
it in that way. So now we're we're this is
56:47
already over an hour, which I'm excited about.
56:49
This has been a great conversation that you did talk
56:52
to me on one of our last calls about open claw and I think
56:56
there's some really, really cool, fascinating and a future
56:58
forward things there. Talked a little bit about people
57:01
getting their own computers, making sure that as security is
57:04
important, put this on an isolated system.
57:07
But can you share a little bit about like what that is and what
57:11
people are trying to do and how that in many ways is probably
57:14
like a new frontier for AI use today?
57:19
We're talking about Njentic and we're talking about these agents
57:21
being able to do tasks, right? I gave you an example of the
57:25
long research task, right? So what other long tasks that I
57:29
gave an AI? And can I have AI take multiple
57:32
actions on my half? Can my AI even run a company on
57:35
my behalf? So those are questions that
57:38
people are having and people are experimenting within Open Claw.
57:42
How I use Open Claw is a little, I'm not necessarily at that
57:45
advanced level, and I've used it as a personal agent.
57:48
So to give you an example of how I'm using it right now is that I
57:52
have AI have a with the chief of staff.
57:55
His name is Rufus. Yeah.
57:58
And then I have different agents.
58:00
Now, if you watch any anime, try to see if you can actually guess
58:04
what. Guess what might these agents
58:05
do? My first my one agent's name is
58:08
Sybil. My other agent is named Lloyd.
58:11
I have another agent named Kira, and I have another agent named
58:16
Sinku. So Sinku is my research agent.
58:20
Going out there trying to find a lot of different information and
58:22
trying to gather a lot of different things.
58:25
Kira is my critique agent. It kills every idea, challenges
58:29
anything, and goes against whatever The thing is.
58:32
Lloyd is my marketing agent, so looking at different things, be
58:38
it in a brand, be able to copy and be able to do things.
58:41
Sybil is my builder, so that's where I'm looking at AI, being
58:45
able to do things. So one of the things I'm
58:48
experimenting with is is I'm looking at research papers that
58:51
were done at a conference called Lean Construct.
58:57
It's IGLC, which is a international global lean
59:01
construction and they have a lot of research papers that are out.
59:04
So when I'm doing is I'm looking at the themes of the.
59:07
So one of the things I'm doing is looking at the themes of the
59:09
construction papers and actually analyzing those themes and
59:13
looking at them from from that standpoint.
59:14
My agent is doing that. It's going through reading all
59:17
the stuff, mapping it together and actually coming out with the
59:20
things for that. But for the past three years,
59:22
Chechi PT or AI has been a thing, right?
59:25
So what are the things that people are talking about?
59:28
What are the things that people are are actually pressed right
59:30
now and can I build applications based off of the the scientific
59:34
research that people have have worked on?
59:37
A lot of our applications, a lot of our ideas and a lot of our
59:39
things are based off of certain somebody in Africa said this was
59:44
a problem and somebody said I'm going to build the solution.
59:48
So So what I'm doing is it's actually using open claw to go
59:53
ahead and look at those solutions and actually build out
59:56
applications while I sleep. I'm not even touching it, not
59:58
looking at it, not even doing anything.
1:00:00
It curbs up the plan, looks at the plan, creates the plan, and
1:00:04
then trying to build out that plan.
1:00:05
So that's that is one thing that I'm trying to do.
1:00:08
So the the level of content and the more work that I'm and I've
1:00:13
never owned, never ran, never operated a software company, but
1:00:17
these are the things that I'm trying to show is like potential
1:00:19
as far as tools that are out there.
1:00:21
One of the things that came out, I know.
1:00:24
Here's here's another example. The government has the big
1:00:27
beautiful bill. That was a big thing for a lot
1:00:29
of people. And then the previous
1:00:31
administration before that was the Biden infrastructure plan
1:00:34
for construction, right? So they were looking at that or
1:00:37
trying to figure out what are the things, what are the ideas,
1:00:39
what are the things I can go ahead and be able to take
1:00:42
advantage of? Well, now I can throw a, now I
1:00:44
can throw an AI agent at it. Because most of the time those
1:00:48
legal, those bills that these Congress people write, because
1:00:52
I've done the homework and I've actually digged into it, are
1:00:54
usually not about what you think, what the media is talking
1:00:57
about. There's a lot of things that are
1:00:58
thrown in. There is a lot of noise, a lot
1:01:00
of things to be able to sift through being by throwing an AI
1:01:03
agent at it. Being able to look at all these
1:01:05
different angles and different situations, how you could take
1:01:07
competitive advantage on certain things would be advantageous for
1:01:10
a company to be able to do. Yeah, this is like, I love
1:01:14
hearing about it because it's, it's so exciting to think about.
1:01:17
Like you're building things while you sleep because you've
1:01:20
got these agents you were trained up on how you would go
1:01:24
about doing that specific task and then getting them all kind
1:01:28
of orchestrated together with Rufus there.
1:01:31
I think it was just fascinating. So we always end these episodes
1:01:35
with one one question. And I'm trying to figure out how
1:01:38
to put some context around it because we definitely know that
1:01:41
you're a wire to build the kind of guy.
1:01:43
But I guess in in many ways less like what?
1:01:46
What would you be going to to keep learning and keep sharp
1:01:51
with this? Like what helps you stay wired
1:01:53
to build? I think the thing that keeps me
1:01:55
wired right. And, or keeps me excited right
1:01:58
now is the built future that I, I could imagine as far as like
1:02:04
integrated society or society being more integrated.
1:02:08
It's starting to play out and I just want to be a contributor to
1:02:12
that world, to that to that, to that space.
1:02:15
You're going to see a lot more smart cities become smarter.
1:02:19
And I think when it really kicks off is when we start seeing
1:02:24
physical AI play out in the world instead of seeing homes,
1:02:29
maybe even reimagined where we look at and say, hey, let's
1:02:34
build a, let's have a robot in our house, right?
1:02:36
To help me clean, wash clothes, do all the little chores and
1:02:39
things that we don't want to do, right?
1:02:41
Well, maybe the way that we have our houses built right now may
1:02:45
not be advantageous to that robot.
1:02:48
Maybe we have to start rethinking how we build our
1:02:51
homes to fit the robot. So those type of questions and
1:02:55
those type of things and those type of things that like kind of
1:02:58
like keep me up or self driving cars.
1:03:02
Well, self driving cars and we may have a lane for just self
1:03:06
driving cars where it can just you have a lane in your city
1:03:11
where the self driving car ends up going down.
1:03:14
So they can do different things that other cars cannot do right.
1:03:19
They can go at certain speeds and it could be a certain
1:03:21
distance away from each other. And then that's the lane that
1:03:24
these self carrying cars couldn't kink to be on.
1:03:27
I often joke about the world where, you know, I'm a old man
1:03:30
and a young kid goes, you change the lane by yourself.
1:03:35
I'm I'm looking for I can't wait for that.
1:03:38
Your answer is try to buy a whole other hour of this pocket.
1:03:41
It's 'cause I'm like trying so hard, like, yeah, let's do this.
1:03:45
We're going to have to invite you back because yeah, I, I
1:03:48
would always, I can't avoid getting out of the soapbox.
1:03:50
So like, we could have automated cars tomorrow if we just got
1:03:53
people off the road because we're the random factor here.
1:03:56
And the dedicated lane, all of that stuff makes total sense.
1:04:00
And then spilling that over into like, well, how does the house
1:04:03
change itself? Like there's, there's
1:04:06
significant infrastructure evolution that is going to occur
1:04:13
and it is going to be really funny to look back, I think and
1:04:17
think of the things that our our walk uphill both ways in snow is
1:04:22
going to is going to seem. Insurmountable.
1:04:27
They're going to be like you did what?
1:04:29
You had to open your car door and drive it.
1:04:33
You had to pay attention the whole time.
1:04:35
What take a nap for this long ride didn't even?
1:04:39
Need to because he went 300. MPH and you were there in 2
1:04:42
minutes. Thank you so much for joining
1:04:44
the show. I have some such a great time.
1:04:47
I'm excited for our next call. Or we're not even recording.
1:04:50
We're just catching up over what we're up to.
1:04:52
I really appreciate you being on.
1:04:54
Yeah, definitely. It was fun.
1:04:55
This was fun. Thank you.
1:04:56
Thank you. You.
1:04:57
Actually asked some questions that I didn't have to stop and
1:04:59
think about and I was like, oh, OK, this is how how I, I, I
1:05:04
enjoyed the way that you be. You were able to pull things for
1:05:07
me that I didn't even necessarily realize.
1:05:08
So thank you for that. One of the things that stands
1:05:11
out from today's conversation. Is that AI really isn't about
1:05:15
replacing builders, it's about amplifying the people who are
1:05:18
willing to explore new tools and understand how they fit into the
1:05:21
real work. What I appreciate about your
1:05:24
perspective, Marcus, is that it is grounded in experimentation.
1:05:27
Not theory, not hype, just figuring out how these tools can
1:05:32
actually help them do their jobs better.
1:05:34
Really appreciate you taking the time to share what you're
1:05:37
seeing. If you enjoyed this episode,
1:05:40
make sure you follow Wired to Build wherever you listen and
1:05:43
share it with someone working at the intersection of
1:05:46
construction, operations, and technology.
1:05:49
You can also find clips full of episodes on LinkedIn with more
1:05:53
Season 3 conversations on the way.
1:05:55
Thanks for listening and until next time, stay Wired to Build.